Claude Mythos 5.1

// Overview

Claude Mythos 5.1 is the restricted-access half of Anthropic's 1 September 2026 Mythos-class release. It is the same underlying model as Claude Fable 5.1, carrying the same specifications and pricing, but ships with lighter safeguards aimed at work in cybersecurity and the life sciences. Anthropic offers it by invitation only through Project Glasswing, limited to US organizations in its Cyber Verification Program and Life Sciences Verification Program; access runs through an Anthropic, AWS or Google Cloud account team.

Because it shares the Claude Fable 5.1 model, Claude Mythos 5.1 has a 1M-token context window, up to 128K output tokens, text and image input, and adaptive thinking that is always on with a default effort of high. Pricing matches Claude Fable 5.1 at $10 per million input tokens and $50 per million output tokens, with cache reads at $0.25 per million.

Anthropic reports that the lighter safeguards change how often the model declines benign work: biology safeguards trigger 85% less frequently on benign medical queries, and cybersecurity safeguards show 60% fewer false positives per session. Anthropic also states that the model can identify software vulnerabilities for defensive work.

Anthropic published several scientific results for Claude Mythos 5.1. On three targets it designed protein binders whose binding affinities were 10 times higher than the best designs submitted to Adaptyv Bio's protein design competitions, with a hit rate of nearly 50% across 12 targets. It produced a Venus elevation map resolving details down to two to three kilometres rather than 10 to 20, with heights up to 25% more accurate. Using custom GPU kernels and cached intermediate results, it sped up seven open-source deep learning models by up to 2.5 times with identical outputs. On Terminal-Bench 4.0, Anthropic reports Claude Mythos 5.1 at 60.9%, above the 55.8% it reports for Claude Fable 5.1.

// Benchmarks

1. Terminal-Bench 4.0 (agentic coding)60.9%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Shares Claude Fable 5.1's specifications and pricing.

Pricing source ↗

// Strengths

• Same capability as Claude Fable 5.1 with safeguards tuned for cybersecurity and life-sciences work
• 60.9% on Terminal-Bench 4.0, above the 55.8% Anthropic reports for Claude Fable 5.1
• Biology safeguards trigger 85% less often on benign medical queries; cybersecurity false positives down 60% per session
• Demonstrated molecular design results — protein binders with 10x higher affinity than the best competition entries on three targets
• 1M-token context window and 128K max output, at Claude Fable 5.1's pricing

// Best for

• Run defensive cybersecurity work, including identifying software vulnerabilities, inside a verified organization.
• Support life-sciences research such as molecular and protein binder design where general-availability safeguards would decline benign requests.
• Take on long-horizon agentic engineering, including writing custom GPU kernels to speed up existing models.
• Use it as a drop-in for Claude Fable 5.1 when your verified programme grants access and safeguard false positives are the blocker.

// Claude Fable / Mythos (Mythos-class) — every version

The full lineage of the Claude Fable / Mythos (Mythos-class) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.